Langfuse Tracing for AI Operations

Available Modules

_shared/langfuse/
├── shared.ts      # Base: Tracer, withTrace()
├── gemini.ts      # Gemini: text, tools, images
├── anthropic.ts   # Anthropic: Claude text, tools
└── openai.ts      # OpenAI: text, tools, images

Basic Usage

Wrap Function with Tracing

import { withTrace, traceLLMGemini } from '../_shared/langfuse/gemini.ts';

Deno.serve(async (req: Request) => {
  return await withTrace(
    {
      name: 'my-function',
      userId: user.id,
      sessionId: item_id,
      input: { prompt: 'analyze this' },
      metadata: { function: 'my-function' },
      tags: ['ai', 'analysis'],
    },
    async (tracer) => {
      // Your AI calls here
      const { text } = await traceLLMGemini(tracer, {
        model: 'gemini-2.5-flash',
        messages: [{ role: 'user', content: 'Hello' }],
      });
      
      return new Response(JSON.stringify({ result: text }));
    }
  );
});

Gemini: Text Completion

const { text, usage } = await traceLLMGemini(tracer, {
  model: 'gemini-2.5-flash',
  messages: [{ role: 'user', content: 'Summarize this' }],
  temperature: 0.7,
  maxTokens: 500,
});

Gemini: Structured Output

const { data } = await traceLLMGemini<{ title: string; price: number }>(tracer, {
  model: 'gemini-2.5-flash',
  messages: [{ role: 'user', content: 'Extract product info' }],
  schema: {
    type: 'object',
    properties: {
      title: { type: 'string' },
      price: { type: 'number' },
    },
    required: ['title', 'price'],
  },
});
// data.title, data.price are typed!

Gemini: Function Calling

const result = await traceLLMWithToolsGemini(tracer, 'agent-name', {
  model: 'gemini-2.5-flash',
  messages: [{ role: 'user', content: 'Get weather in SF' }],
  tools: [{
    name: 'get_weather',
    description: 'Get current weather',
    parameters: {
      type: 'object',
      properties: { location: { type: 'string' } },
      required: ['location'],
    },
  }],
});

if (result.message.tool_calls) {
  // Execute tools
}

What Gets Tracked

✅ Traces: Function execution with nested operations ✅ Generations: LLM calls with prompts and responses ✅ Usage: Token counts and costs ✅ Latency: Timing for each operation ✅ Metadata: Custom tags and context ✅ Errors: Automatic error tracking

Setup

Set environment variables in .env.local:

LANGFUSE_PUBLIC_KEY=pk-lf-xxxxx
LANGFUSE_SECRET_KEY=sk-lf-xxxxx
LANGFUSE_HOST=https://cloud.langfuse.com

Available Functions

Gemini: traceLLMGemini, streamLLMGemini, traceLLMWithToolsGemini, traceImageEditGemini Anthropic: traceLLMAnthropic, streamLLMAnthropic, traceLLMWithToolsAnthropic OpenAI: traceLLMOpenAI, streamLLMOpenAI, traceLLMWithToolsOpenAI, traceImageEditOpenAI